A disruption predictor based on a 1.5-dimensional convolutional neural network in HL-2A

Abstract

Disruption means a sudden loss of confinement during a discharge in fusion reactors. Due to the huge electromagnetic loading and thermal loading on the facility and a large number of runaway electrons generated during disruptions, it is essential to find a method to predict the disruptions, so that measures like massive gas injection can be taken to mitigate or to avoid these harmful effects. In this research, a machine learning model mainly based on a 1.5-dimensional convolutional neural network, which is good at dealing with signals from multi-channels with great divergence, is trained to predict disruptions in the HL-2A tokamak. The disruption predictor uses shots 20000–29999 in HL-2A to train the machine learning model, and uses shots 30000–31999 to optimize hyper parameters. When tested on shots 32000–36000 in HL-2A, it reaches a true positive rate of 92.2% and a true negative rate of 97.5% with 30 ms before the disruption.